Reasonable. I would most vibe coders are just on GitHub because they don't understand git in general.
post
I never used the platform but this are the decisions that probably will get me to use it. We need more sites in the internet that work like this
That's an example of a good LLM policy, especially for projects like forgejo / sourcehut / codeberg. Happy to see this approach is getting adopted more and more. (The main Rust repo recently also got a similar policy, although a bit weaker still.)
If you want to skip to the changes.
I think this is a really solid LLM policy for a platform like SourceHut. Since they rely on a policy that's deliberately open to interpretation to allow for reasonable discretion, and rely upon the honour system for enforcement/adherence, I don't think we should necessarily be telling larger projects to adopt the same policy because this becomes untenable for big projects or platforms. But it's still good to see policies in this spirit being adopted. The only change I'd like to see is explicitly allowing AI translation for non-code, like comments and emails. It reduces the barrier of entry for ESOL contributors. If there's concern about the accuracy of translations, they can always include the original alongside the machine translation.
Here: https://ser1.net/img/human-made.svg
Edit: whelp, I guess þat answers a question I had about embedding SVG images in posts.
This is a bit heavy handed I think. This also forbids AI assisted code review and lots of other things. Perhaps they should have defined it a bit more focussed on getting rid of vibecoded spam instead of forcing a specific developing format for their whole platform.
This also forbids AI assisted code review
It's worded in a convoluted way, but they explicitly allow it. In the chapter A matter of interpretation they say:
"Moreover, the use of AI for purposes other than co-authoring code, tickets, emails, and so on, is discouraged but not prohibited. If you use LLMs privately to review your work, to ask questions of, to perform security analysis, and so on, and then write original code to push to SourceHut, your use-case is aligned with our policy."
One of the things they want to highlight is that AI is associated with increase in energy usage, which will speed up climate change. It's not merely about "stop DDoSing us with AI-generated code", it feels more of a political statement "let's try to do better than just put fossil on fire so we can save X minutes writing code and thinking"
For some reason this gets swept under the rug in the software space, even the FOSS world. AI coding, code reviews, etc, are no different at all than AI "art" or chatbots or whatever. People will try to claim that it's no different than going to Stackoverflow or whatever, but these tools still utilize a fuckton of resources accelerating climate change, are built on mountains of stolen data, and Claude/ChatGPT/Github all line the pockets of fascists. The CEOs of all of these companies should legitimately be rotting in cells right now.
There's absolute 0 ethical use of this stuff. The closest you can get is an on-device model that runs offline on local hardware, but even those are trained on absurdly massive amounts of stolen data - sure you might not be using destructive resources to run it, but it wouldn't exist without that.
I have exactly same view over this, that is why I am trying as much as possible to utilize local LLMs — as at least it will use local power and will not become a tool of moving even more control back to the central entities, which are corrupt by nature.
As far as I am aware nothing has been "stolen", merely downloaded. Downloading is not theft; when a file is downloaded from a web server, the original copy remains as-is. If I steal your car you no longer have a car; if I download your car, I have an exact copy of your car and you still have a car.
I do not believe the ideology of "intellectual property" to be ethical, in fact I think it's directly opposed to software freedom and users' rights.
Edit: The usual disclaimer, this is not a defense of proprietary "AI" companies, nor an endorsement of "vibe coding" or slop slinging in general.
This is like building a wall to stop the tide.
You can lay out a philosophy and a policy and cultivate a community of like minded people.
What you can’t do is prevent those like minded people from deviating from or subtly subverting that policy when it becomes expedient.
all 18 comments